IC-P2-099: Systematics of the ADNI phantom

Abstract

The Alzheimer's Disease Neuroimaging Initiative (ADNI) employs a fleet of 66 phantoms (photo below) for MRI scanner characterization and to insure the integrity of data pre-processing steps. Eighty-four different scanners are used in ADNI. The ADNI phantom is scanned along with the subject for every ADNI MRI exam. Geometric fidelity is assessed by comparing the design with observed locations of 160 small spherical fiducial markers. Differences in expected and observed positions are plotted versus observed positions below. SNR and relative contrast are estimated from larger spheres containing solutions of known copper sulfate concentration. Phantom images are acquired with the same MP-RAGE pulse sequence used for human imaging. The phantom scan analysis is fully automated and provides estimates of gradient calibration (proportional to an object's size as measured by MRI), non-linear geometric distortion, SNR and contrast properties. Estimates of within phantom measurement stability were obtained by repeated scanning. Estimates of phantom to phantom variability were obtained by comparing all phantoms to single phantom. Repeatability was assessed by scanning a single phantom multiple times; between scans the phantom was removed and replaced. Coefficients of variation (CV) for SNR and contrast measures were found in the 2–4% range. Gradient calibration CV's were in the range of 3x10–4 to 5x10–4, the larger CV in the scan frequency encoding direction. The phantom construction is a stacked plate design. Phantom to phantom variation is smallest parallel to the plates. Observed CV's parallel to the plate are essentially the same as the within phantom measurement error (i.e. within-plate the manufacturing variability is likely smaller than the measurement error). The CV perpendicular to the plate is 15x10–4 (3X to 5X larger than measurement error), and is assumed to be due to variation in the plate stacking during manufacturing. The leading source of potential error in gradient measurements using the ADNI phantom is phantom-to-phantom variability. Owing to the phantom design, that variability is anisotropic. Given a multi-year study, with thousands of MRI studies, phantoms will occasionally be damaged (dropped) and must be replaced introducing a confound in the consistency of calibration measurements.